On connection from compute, wait for tenant to become Active.

If a connection from compute arrives while a tenant is still in
Loading state, wait for it to become Active instead of throwing an
error to the client. This should fix the errors from test_gc_cutoff
test that repeatedly restarts the pageserver and immediately tries to
connect to it.
This commit is contained in:
Heikki Linnakangas
2022-11-19 11:49:35 +02:00
committed by Christian Schwarz
parent 7552e2d25f
commit e8db20eb26
3 changed files with 66 additions and 12 deletions

View File

@@ -56,9 +56,24 @@ def test_pageserver_restart(neon_env_builder: NeonEnvBuilder):
cur.execute("SELECT count(*) FROM foo")
assert cur.fetchone() == (100000,)
# Stop the page server by force, and restart it
# Restart the server again, but delay the loading of tenants, and test what the
# pageserver does if a compute node connects and sends a request for the tenant
# while it's still in Loading state. (It waits for the loading to finish, and then
# processes the request.)
env.pageserver.stop()
env.pageserver.start()
env.pageserver.start(extra_env_vars={"FAILPOINTS": "before-loading-tenant=return(5000)"})
# Check that it's in Loading state
client = env.pageserver.http_client()
tenant_status = client.tenant_status(env.initial_tenant)
log.info("Tenant status : %s", tenant_status)
assert tenant_status["state"] == "Loading"
# Try to read. This waits until the loading finishes, and then return normally.
pg_conn = pg.connect()
cur = pg_conn.cursor()
cur.execute("SELECT count(*) FROM foo")
assert cur.fetchone() == (100000,)
# Test that repeatedly kills and restarts the page server, while the